ROADMAP

From Claude Code plugin to multi-surface platform.

CLAUDE CODElive
DESKTOPv3.0
COWORKv3.0
EVERYWHEREv5.0
v1.0Think It Through
Claude CodePhases 1-5, 20 plans
SHIPPED March 22, 2026

When I have a venture idea, help me think it through systematically so I don't waste months on the wrong problem.

Solo founder with an idea. No team yet. Needs structured thinking, not another chatbot.

Larry as your thinking partner
26 PWS methodology commands
Auto-organizing Data Room with 8 sections
Knowledge graph dashboard
PDF exports for investors
Pipeline chaining across frameworks
Optional Brain MCP (21K nodes)
v2.0Capture Every Meeting
Claude CodePhases 6-9, 13 plans
SHIPPED March 24, 2026

When my team has meetings, capture what was said so insights don't get lost and patterns emerge across conversations.

Now you have a team. Meetings happen. Decisions scatter. Assumptions go stale between conversations.

File meeting transcripts (paste, file, or audio)
Speaker identification with 12-role profiles
Cross-meeting convergence and contradiction detection
Action item tracking across meeting history
Team intelligence with expertise gap analysis
Auto-fetch from Read AI, Vexa, or Recall.ai
Knowledge graph with meeting and speaker nodes
v3.0MCP Platform
Claude Code + Claude Desktop + CoworkPhases 10-15, 12 plans
SHIPPED March 25, 2026

When I use Claude Desktop or Cowork, give me the same intelligence I get in Claude Code - plus new capabilities that need a server.

MindrianOS runs on every Claude surface. 6 phases shipped in one day: MCP server, Brain hosting, Opportunity Bank, Funding Room, AI Personas, and embedded knowledge graph. Two levels of intelligence: Brain for methodology, Room Graph for your venture data.

MCP server — 49 tools routing all 50+ commands
Brain MCP server with API key auth
/mos: prefix + thinking traces + visual confirmations
Opportunity Bank — context-driven grant discovery
Funding Room — 4-stage lifecycle tracking
AI Team Personas — De Bono Six Hats perspective lenses
Embedded knowledge graph — zero-server, local
Two-graph architecture (Brain + Room Graph)
v4.0Brain API & CLI UI
Claude Code + Claude Desktop + CoworkPhases 20-25, 12 plans
SHIPPED March 29, 2026

When I share the Brain with others, control who gets access. When I use any command, see a consistent, professional visual language.

The Brain is the moat. v4.0 locks it down with API key management, then establishes the visual grammar every MindrianOS interaction follows. Multi-room lets you work on multiple ventures. The autonomous engine lets Larry think independently.

Brain API key management (Supabase, approve/revoke/extend)
CLI UI Ruling System (4-zone anatomy, 5 body shapes, 12 glyphs)
Admin panel (/mos:admin, self-teaching, hidden)
Multi-room management (/mos:rooms, context lock, header canary)
Autonomous engine (/mos:act, Brain-driven framework selection, chain mode)
De Stijl HTML export (Mondrian grid, document reader, intelligence, Cytoscape graph)
v5.0Presentation System
CLI + Vercel + Any Static HostPhases 26-32, 12 plans
SHIPPED March 31, 2026

When I need to share my Data Room, generate a professional 6-view site that anyone can explore - with a living knowledge graph and chat with Larry.

The Data Room becomes visible. One command generates 6 self-contained views. A Canvas knowledge graph with particles and glow. Chat with Larry using your own API key. Deploy to Vercel with one click. Every filing auto-updates everything.

6-View Presentation (Dashboard, Wiki, Deck, Insights, Diagrams, Graph)
Canvas knowledge graph with particles, glow, and cluster highlights
BYOAPI chat - Larry generates views and manipulates the graph
Git integration - optional room version control
KuzuDB automatic backbone on every filing
HSI pipeline - surprising cross-artifact connection discovery
Binary asset filing (PDFs, images, videos with manifests)
Vercel one-click deploy with privacy controls
Dual themes: De Stijl dark + PWS light
v1.6.0Powerhouse
Claude Code + Claude Desktop + CoworkPhases 33-40, 8 plans
SHIPPED March 31, 2026

When I have multiple agents, models, and rooms running, give me the controls to orchestrate them intelligently -- routing, parallelism, monitoring, and future-proofing.

The platform matures. Model routing lets each agent pick the best model for its job. Six new lifecycle hooks give plugins real-time awareness. Parallel agent execution unlocks swarm-mode thinking. Spectral analysis reveals how Claude actually reasons. Design-by-Analogy brings TRIZ and SAPPhIRE to venture innovation. Sentinel Intelligence watches your rooms while you sleep.

Model Routing -- per-agent model selection with venture-stage adaptive hints
Hook Expansion -- 6 new hooks (PreCompact, PostCompact, FileChanged, CwdChanged, SubagentStop, TaskCompleted)
Parallel Agents -- --swarm, --parallel, --full, --broad flags
Spectral OM-HMM -- Markov chain thinking-mode analysis
Design-by-Analogy -- 5-stage pipeline with TRIZ + SAPPhIRE
Sentinel Intelligence -- /mos:scout scheduled room monitoring
Platform Optimization -- prompt cache, @include CLAUDE.md, deep links
Future-Proofing -- KAIROS prep, Coordinator Mode, MWP specification
v6.2RoomHub + SnapshotHub
Claude Code + Claude Desktop + CoworkPhases 41-46, 6 plans
SHIPPED March 31, 2026

When I want to see, share, or explore my Room as a living intelligence hub, give me adaptive views, a constellation graph, and a chat that queries my knowledge graph in natural language.

The Room becomes a living, adaptive intelligence hub. RoomHub detects your project type (venture/website/research/general) and tailors everything. SnapshotHub exports 7 views as standalone HTML. A 12-thread Constellation graph with De Stijl spectral coloring makes your knowledge graph navigable. Fabric Chat lets you query KuzuDB in natural language. And JTBD-powered contextual discovery surfaces the next command you need every 3-7 turns.

RoomHub -- living adaptive intelligence hub for any Room
SnapshotHub -- /mos:snapshot exports 7-view standalone HTML
7 Views: Overview, Library, Narrative, Synthesis, Blueprint, Constellation, Chat
Adaptive Room detection (venture/website/research/general)
12-Thread Constellation graph with De Stijl spectral coloring
Generative Fabric Chat querying KuzuDB via natural language
JTBD-powered contextual command discovery every 3-7 turns
New command: /mos:snapshot (63 total commands)
v1.9.0Whitespace Mapping
Claude Code + Claude Desktop + CoworkPhases 60-66 (7 phases, 22 requirements)
SHIPPED April 8, 2026

When I need to find what's MISSING in my venture understanding -- not what connects or what's bottlenecked, but what nobody has thought of yet -- map the empty spaces in my knowledge and tell me what should be there.

The breakthrough release. Based on Huan He's SemNovel research (Yale), MindrianOS now embeds every room artifact and every Brain framework into the same mathematical space, then finds the empty regions -- the whitespace. But finding gaps is commodity math. What makes this ours: the Brain classifies each gap by problem type, selects the right framework chain, and generates methodology-aware hypotheses. That interpretation layer -- teaching intelligence applied to embedding-space gap detection -- is potentially patentable. The Discovery Cycle chains HSI, Whitespace, Reverse Salient, and Analogy into a single pipeline that finds what humans miss. The Brain learns from every room's discoveries.

/mos:whitespace command -- 7 subcommands: map, analyze, hypothesis, tree, score, external, discover
SemNovel novelty scoring -- every artifact scored by embedding distance from Brain consensus
Per-section WHITESPACE.md -- ICM-native gap files in every room folder
Interpretation Layer (the moat) -- problem type classification + framework chain selection
Three-gate validation -- Anchor, Brain Consensus, Semantic Coherence gates filter false positives
TopicForest hierarchical clustering -- topic tree shows coverage gaps at multiple granularity levels
Discovery Cycle -- HSI -> Whitespace -> RS -> Analogy chained pipeline
Brain Intelligence Layer -- cross-room whitespace learning with anonymized patterns
Semantic Scholar integration -- query 200M+ papers for cross-domain whitespace
D3.js density map + TopicForest tree in De Stijl dashboard
KuzuDB WhitespaceZone nodes with full graph persistence
~7,500 lines of whitespace engine code, 79+ tests
v1.8.8Brain Optimization
Claude Code + Claude Desktop + CoworkPhases 60-64 (5 phases, 28 requirements)
SHIPPED April 7, 2026

When I ask Larry to recommend a framework, he should know exactly which one fits my venture stage, what chains to next, and why -- grounded in the source books, not guessing.

The Brain graph went from 4 real framework chains to 125. From 4 stage mappings to 129. From zero prerequisite edges to 15. 444 semantic bridges now connect the LazyGraph layer (245K CO_OCCURS edges) to curated canonical nodes. 235 concepts promoted. 75 noise books cleaned. Every case study wired. Seven leadership books codified with full provenance chains. The teaching graph now knows not just WHAT to recommend, but WHEN in the venture journey each framework matters most -- and WHY, grounded in source books. Plus: every script error is now human-readable, and the install guide walks non-technical users through setup in 10 minutes.

Brain: FEEDS_INTO 4->125 framework chains including full PWS spine
Brain: TYPICAL_AT 4->129 venture stage mappings with effectiveness scores
Brain: PREREQUISITE 0->15 edges enabling 'do X before Y' warnings
Brain: 444 LazyGraph-to-canonical bridges via ALIAS_OF
Brain: 235 high-value concepts promoted from LazyGraph to Concept
Brain: 7 leadership books codified (Golden Circle, Level 5, Hedgehog, Flywheel, etc.)
Brain: 29/29 CaseStudies wired, 406 TEACHES, 23 IMPLEMENTS edges
Brain: Fragmentation cleanup -- 75 noise books removed, 12 label types fixed
Dummy-proof install: human-readable error messages (What/Why/Fix pattern)
Install guide with 3 MECE paths (no Claude / has Claude / update)
Top 10 troubleshooting items on website install page
Fresh install test checklist for Mac and Windows
v1.8.4Cowork Adaptation
Claude Code + Claude Desktop + CoworkPhases 52-61 (10 phases, 61 requirements)
SHIPPED April 5, 2026

When I install MindrianOS, it should work on CLI, Desktop, AND Cowork -- same commands, same intelligence, same Data Room. Brain-driven routing picks my frameworks. Scheduled intelligence watches while I sleep.

The biggest architectural shift since v1.0. MindrianOS now works identically across all three Claude surfaces. 64 MCP tools through 9 hierarchical routers. Brain-driven routing recommends framework chains with 3-tier fallback. Skills compressed 65% via native-first architecture. Scheduled intelligence scans competitors, grants, and news. 6 De Bono hats maintain cross-session memory. Interactive dashboard, wiki, and graph render inline via MCP Apps. Token cost halved from 20K to 10K per turn.

64 MCP tools across 9 hierarchical routers -- every command works on Desktop and Cowork
Brain-driven routing with 3-tier fallback (cache -> local heuristic -> Brain API with 2s timeout)
Intelligence cascade shared module -- MCP writes trigger same HSI/cross-ref/graph as CLI hooks
Surface auto-detection (CLI/Desktop/Cowork) with dual transport (stdio + Streamable HTTP)
KuzuDB write-gateway with promise-chain serialization preventing concurrent write corruption
Skills compressed 65% (74K -> 25.5K bytes) via native-first architecture
Progressive 3-layer skill loading -- per-turn tokens halved (20K -> 10K)
User archetype detection (venturist/researcher/student) with tiered context loading
MCP session profiles (learn/think/build/research/present/full) control which servers load
Pipeline chaining via room artifacts -- scenario -> root-cause -> causal -> prediction end-to-end
Budget-aware agent dispatch with cost estimation and chain checkpointing
Scheduled intelligence: daily briefings, competitor/grant/news scanning on Cowork
De Bono persistent hats with cross-session memory feeding Brain routing
3 MCP Apps: dashboard, wiki, knowledge graph rendered inline in Desktop/Cowork
KAIROS/Coordinator/UDS platform readiness stubs for upcoming Anthropic features
66 total commands, 64 MCP-routed (hat-briefing and scheduled-tasks added)
v1.7.1Causality Engine
Claude Code + Claude Desktop + CoworkPhases 52-57 (Causal) + Quick Tasks (Beta Fixes)
SHIPPED April 5, 2026

When I need to understand WHY something works in my venture, trace causal chains with mechanisms. When I export my Room, get a production-quality shareable hub.

The biggest release since v1.0. Causal Reasoning Layer adds 5 new commands for tracing cause-effect chains with mechanisms, falsifiable predictions, natural experiments, DAG diagrams, and reasoning audits. The export system was rebuilt from scratch -- generate-hub.cjs produces single-file De Stijl hubs matching the Synteris production template. 15 critical bugs fixed from the first external beta test. Full macOS compatibility. And MindrianOS now manages itself -- its own Data Room with 63 artifacts, 19 decisions, 14 wishes, and a self-referential evolution loop.

Causal Reasoning -- /mos:causal extract, validate, diagram, what-if, audit
Synteris-quality hub export -- single-file De Stijl with venture cards, grade circles, badges
Smart content detection -- bug/wish/decision cards with status badges
15 beta fixes -- all 29 commands unblocked, Brain setup, statusline, onboarding
macOS compatibility -- Python3 fallbacks for bash 3.2, POSIX grep alternatives
MindrianOS Data Room -- the product manages its own venture as a case study
Product evolution loop -- wishes that become features, patterns that reveal priorities
Recursive scanning -- all visualization scripts handle nested directory structures
68 total commands (5 new causal reasoning commands)
v1.8.6MindrianRooms
Claude Code + Claude Desktop + CoworkPhases 56-59.2 (6 phases, 35 requirements)
SHIPPED April 6, 2026

When I have multiple ventures, organize them in one place where the folder structure itself holds understanding -- and let the graph discover why rooms belong together.

Room organization is a wicked problem. A static folder hierarchy forces one axis -- by client, by stage, by domain -- but ventures exist on all axes simultaneously. v1.8.6 centralizes rooms under ~/MindrianRooms/ with ICM-compliant identity files at every level, then adds a dual-graph intelligence layer (KuzuDB local + Neo4j Brain remote) that holds ALL organizational axes as a graph. The /mos:organize command navigates this wicked hierarchy with virtual projections, graph-informed grouping proposals, and human confirmation for every move. Decisions feed back into the graph, making each reorganization smarter than the last. Simon's near-decomposable hierarchies, implemented as software.

MindrianRooms centralized directory with ICM Layer 0/1 auto-generation
4-strategy resolve-room cascade (central registry, dir scan, workspace, legacy fallback)
MINDRIAN_ROOMS_HOME env var for power users
/mos:organize -- wicked hierarchy navigator with 4 subcommands (tree/propose/view/move)
Virtual room projections (by-stage, by-client, by-domain) without moving files
Graph-informed grouping proposals with 4-tier graceful degradation
Human-in-the-loop confirmation for every room move
Decision memory -- GROUP/SEPARATE/DEFER choices become graph edges
Dual-graph room hierarchy: KuzuDB local + Neo4j Brain remote (additive, never blocking)
13 orphaned DataRoomSection nodes wired to Room parents in Brain
Migration engine detecting 5 legacy room patterns with guided confirmation
/mos:setup rooms for one-command legacy migration
ICM templates (CLAUDE.md identity + INDEX.md routing) generated from registry
v5.1Proactive Intelligence
CLI + Desktop + Cowork
BACKLOG

When I'm working, Larry should surprise me with insights I wasn't expecting - connections from the graph that change how I think about my venture.

MindrianOS becomes proactive. Larry monitors KuzuDB in the background and surfaces surprising connections mid-conversation. Strategic clarifying questions before every action. De Stijl CLI visual identity.

Proactive KuzuDB insights mid-conversation
Strategic clarifying questions (Larry's discuss-phase)
De Stijl CLI visual identity (branded terminal output)
Decision tracking in nested folder structure

v1.0 teaches you to think. v2.0 teaches your team to think together. v3.0 brings Larry to every Claude surface. v4.0 protects the Brain and lets Larry act autonomously. v5.0 makes the Data Room visible and shareable. v1.6.0 Powerhouse adds model routing and parallel agents. v6.2 RoomHub turns the Room into a living intelligence hub. v1.7.1 adds causal reasoning and production exports. v1.8.4 adapts to all three Claude surfaces with 64 MCP tools. v1.8.6 MindrianRooms treats your venture portfolio as a wicked problem. v1.8.8 Brain Optimization makes the teaching graph 10x smarter -- 125 framework chains, 129 stage mappings, 444 semantic bridges, and every recommendation grounded in source books. Plus dummy-proof installation for non-technical users.